Panoramic Mosaic Of Multi-Cameras Based On GPU

The processing of panoramic image mosaic is mainly described in this paper. We adopt the way of camera calibration to eliminate the lens distortion and to complete the work of image pre-processing considering the characteristics of the new device which is called panorama synchronization camera. This paper uses an improved template matching algorithm to find the similar area of two images. The fist step is to sign up the fusion area roughly.In the second step, we try to make an accurate matching in the fusion area which is found in the fist step. The accuracy of the image registration will be improved according to using this algorithm. Finally, we stitch a panorama with 360°horizontal viewing angle using an image fusion method considering the edge information of the fusion area.The processing of panoramic image mosaic involves bulks of data calculations, which are suitable for using the ability of GPU parallel to complete, such as image projection and image fusion computing operations. Therefore, this paper uses a method of combination of CPU and GPU for data processing, and makes programming under the architecture of CUDA. The result shows that the time that the combined approach costs is far less than that spent by using CPU simply.